Complete coverage path planning for multi-connected free-form surface grinding based on reinforcement learning

Abstract

Coverage path planning (CPP) is an essential process in robotic grinding, particularly with the increasing demand for large-scale multiconnected free-form surfaces, such as high-speed rail shells, car shells, and aeronautical parts. Owing to its multi-connectivity, achieving full coverage with a single continuous path is challenging. Additionally, large curvatures make the path spacing difficult to control, leaving some areas uncovered. Existing methods often fail to optimize continuity and coverage rates simultaneously, resulting in redundant tool-feeding and lifting processes that significantly reduce processing efficiency. Thus, a novel method for free-form surface CPP is proposed based on reinforcement learning (RL), which enables the learning of an optimal path with optimized continuity and coverage rates. Specifically, to regulate the path spacing, a uniform grid map is constructed based on the least-squares conformal mapping (LSCM) method, which parameterizes the grinding surface to a two-dimensional (2D) plane with controllable distortion. Furthermore, a CPP-specific evaluation criteria (CEC) is designed to evaluate the path through various key factors, including coverage rate, continuity, and smoothness. Finally, a grinding path is generated using the CEC-guided RL framework. The method was verified through several simulations, and a grinding experiment on a high-speed rail head surface was conducted as a typical application. The results showed high path continuity and coverage rates, demonstrating its potential for addressing CPP problems in different manufacturing scenarios.

Keywords

Coverage path planning (CPP) / Grinding / Reinforcement learning (RL) / Muti-connected free-form surface / Reward function construction

References

[1] GalceranE, CarrerasM. A survey on coverage path planning for robotics. Robot Auton Syst, 2013, 61(12): 1258-1276

[2] WangH, QiS, SunY, et al.. Study on automatic sanding process of putty for top of high speed railway train body. Paint Coatings Industry, 2023, 53(5): 76-80

[3] HassanM, LiuD. PPCPP: a predator-prey-based approach to adaptive coverage path planning. IEEE Trans Rob, 2020, 36(1): 284-301

[4] WanS, ZhangX, XuM, et al.. Region-adaptive path planning for precision optical polishing with industrial robots. Opt Express, 2018, 261823782.

[5] Nasirian B, Mehrandezh M, Janabi-Sharifi F (2021) Efficient coverage path planning for mobile disinfecting robots using graph-based representation of environment. Front in Robot and AI 8. https://doi.org/10.3389/frobt.2021.624333

[6] Krupke D (2023) Near-optimal coverage path planning with turn costs. arXiv:2310.20340. https://doi.org/10.48550/arxiv.2310.20340

[7] MansorMSA, HindujaS, OwodunniO. Voronoi diagram-based tool path compensations for removing uncut material in 2JD pocket machining. Comput Aided Des, 2006, 38(3): 194-209.

[8] ZouQ, ZhaoJ. Iso-parametric tool-path planning for point clouds. Comput Aided Des, 2013, 45(11): 1459-1468.

[9] SunY, GuoD, ZhangJ, et al.. Iso-parametric tool path generation from triangular meshes for free-form surface machining. Int J Adv Manuf Technol, 2005, 28(7/8): 721-726.

[10] ZouQ, ZhangJ, DengB, et al.. Iso-level tool path planning for free-form surfaces. Comput Aided Des, 2014, 53: 117-125.

[11] FengHY, TengZ. Iso-planar piecewise linear NC tool path generation from discrete measured data points. Comput Aided Des, 2005, 37(1): 55-64.

[12] McGovern S, Xiao J (2002) UV grid generation on 3D freeform surfaces for constrained robotic coverage path planning. In: IEEE 18th international conference on automation science and engineering (CASE), Mexico City, Mexico, 20‒24 August, pp 1503‒1509. https://doi.org/10.1109/case49997.2022.9926608

[13] ShengW, ChenH, XiN, et al.. Tool path planning for compound surfaces in spray forming processes. IEEE Trans Autom Sci Eng, 2005, 2(3): 240-249.

[14] HauthS, RichterichC, GlasmacherL, et al.. Constant cusp toolpath generation in configuration space based on offset curves. Int J Adv Manuf Technol, 2010, 53(1/4): 325-338.

[15] WangS, WangC, WangP, et al.. PDE-based spiral machining trajectory planning method without tool feed marks on 2D arrayed multi-island regions. Int J Adv Manuf Technol, 2023, 125(5/6): 2021-2034.

[16] ChosetH. Coverage for robotics-a survey of recent results. Ann Math Artif Intell, 2001, 31: 113-126.

[17] HuangKC, LianF, ChenC, et al.. A novel solution with rapid Voronoi-based coverage path planning in irregular environment for robotic mowing systems. Int J Intell Robot Appl, 2021, 5(4): 558-575.

[18] KyawPT, PningA, ThuTT, et al.. Coverage path planning for decomposition reconfigurable grid-maps using deep reinforcement learning based travelling salesman problem. IEEE Access, 2020, 8: 225945-225956.

[19] Shen C, Mao S, Xu B et al (2003) Spiral complete coverage path planning based on conformal slit mapping in multi-connected domains. arXiv:2309.10655. https://doi.org/10.48550/arXiv.2309.10655

[20] Holland JH (1992) Adaptation in natural and artificial systems. MIT Press, Cambridge, MA

[21] GrondmanI, BusoniuL, LopesG, et al.. A survey of actor-critic reinforcement learning: standard and natural policy gradients. IEEE Trans Syst Man Cybern, 2012, 42(6): 1291-1307.

[22] Gong Y, Wang L, Guo R et al (2014) Multi-scale orderless pooling of deep convolutional activation features. In: Fleet D, Pajdla T, Schiele B et al (eds) Computer vision – ECCV 2014. Lecture Notes in Computer Science, vol 8695. Springer, Cham. https://doi.org/10.1007/978-3-319-10584-0_26

[23] Schulman J, Wolski F, Dhariwal P et al (2017) Proximal policy optimization algorithms. arXiv:1707.06347. https://doi.org/10.48550/arXiv.1707.06347

[24] HuangS, OntanonS. A closer look at invalid action masking in policy gradient algorithms. Int FLAIRS Conf Proc, 2022.

[25] XingB, WangX, LiuY, et al.. An algorithm of complete coverage path planning for unmanned surface vehicle based on reinforcement learning. J Marine Sci Eng, 2023, 113645.

[26] SuzukiS, AbeK. Topological structural analysis of digitized binary images by border following. Comput Vis Graph Image Process, 1985, 30(1): 32-46.

[27] GrahamRL, YaoF. Finding the convex hull of a simple polygon. J Algorithms, 1983, 4(4): 324-331.

[28] Hexagon (2024) Leica absolute tracker AT960. https://hexagon.com/products/leica-absolute-tracker-at960. Accessed 1 Apr 2024

[29] Hexagon: leica T-Scan 5 (2024). https://hexagon.com/products/leica-t-scan-5. Accessed 1 Apr 2024

[30] PageES, BellmanR. Adaptive control processes: a guided tour. J R Stat Soc Ser A, Gen, 1962, 125(1): 161-162

[31] BotvinickMM. Hierarchical reinforcement learning and decision making. Curr Opin Neurobiol, 2012, 22(6): 956-962.

Funding

Key Technologies Research and Development Program(2022YFB4700501)

National Natural Science Foundation of China(52375495)

Fundamental Research Funds for the Central Universities(23JYCXJJ034)

RIGHTS & PERMISSIONS

Shanghai University and Periodicals Agency of Shanghai University and Springer-Verlag GmbH Germany, part of Springer Nature